This skill helps an agent build the request contract between Roblox clients and the server. The server decides state and clients send intent, so the skill walks through inventorying existing remotes, choosing a transport, minimizing payloads, validating in layers and applying changes on the server.
It includes a combat validation pattern, a token bucket limiter, rules for ranged hit validation and lag policy, guidance for places that use Server Authority, and a symptom table for exploits, throttling and desync.
What You Get
- An eight-step workflow from finding every existing remote to running an abuse matrix with Output from each side
- A table for choosing between reliable events, unreliable events and request and response functions, with delivery rules that affect handler design
- Luau patterns for a validated combat request and a per-player token bucket
- Guidance for Server Authority places and a symptom, cause and remedy table
Common Use Cases
- Auditing a remote that lets the client choose a price or damage value
- Adding rate limits and shape checks to remotes that exploiters spam
- Validating ranged weapon hits on the server with a stated lag policy
- Testing a remote contract with two clients and malformed inputs
What Makes It Unique
The skill starts by listing every remote and both endpoints so a duplicate is not added. It covers yielding handlers that let two requests through, a detail many remote guides miss, and a rule never to call a client from the server. It also says not to switch a place to Server Authority as a side effect of another task.
Under The Hood
The agent loads the SKILL.md and, for contract work, a reference with a worksheet, validation order and the multi-client test matrix. It searches the project source with a command such as ripgrep for remote handlers, edits Luau files, and uses a connected Roblox Studio MCP server to start play, run Luau validators on the server and read the console. Without that server it asks you to run two clients and paste each Output. One metadata file for an OpenAI agent sits beside the skill. Documentation URLs are listed for re-checking APIs, and nothing is fetched during normal use.